AVGO · Broadcom Inc.

arXiv paper proposes UniMVU, an instruction-aware dynamic gating architecture for multimodal video understanding (video+audio+depth/temporal streams). It reduces “modality interference” from uniform fusion by reweighting salient regions within modalities and entire modality streams conditioned on the text instruction, showing sizable benchmark gains. Investable angle: improves accuracy/efficiency of multimodal video agents and sensor/stream fusion, reinforcing demand for GPU/cloud inference and

Post argues co-packaged optics (CPO) and silicon photonics are the next scaling lever for “1M GPU AI factories,” and claims Soitec has a near-monopoly in a critical photonics SOI engineered substrate used across the silicon photonics stack (NVIDIA CPO switches, Broadcom DC ASICs, 800G/1.6T transceivers at hyperscalers). Despite SOI being down ~75%, CEO retiring, and mobile end-market weakness, author expects a multi-bagger as optical interconnect market expands to 2030.

Post argues AI infrastructure bottleneck is shifting from GPUs toward CPUs as agentic/workflow-based AI increases branching, I/O, and decision-heavy tasks. Implies rising CPU demand intensity (CPU:GPU ratio moving toward 1:1) and underappreciated CPU supply/throughput constraints.

Meta says it is expanding its Richland Parish, Louisiana data center to 5GW of compute capacity. The post is largely framed around local economic benefits (teacher bonuses, small businesses), but the investor-relevant signal is the scale of incremental compute/infrastructure buildout, implying sustained AI/data-center capex and upstream demand for accelerators, networking, power and thermal infrastructure.

This is a high-level technical discussion about Cerebras/wafer-scale “dinner plate” computing: compiler complexity, wafer-scale yield, PVT calibration, parallelism approaches, and system bottlenecks (memory and I/O bandwidth), ending with an “economically relevant conclusion.” No explicit public tickers/cashtags, no stated positioning, and no explicit catalyst or valuation call. Actionability is therefore low; it’s mainly context for AI compute hardware economics and bottlenecks.

Podcast summary highlights: accelerating AI capability toward AGI, “race for compute,” effectively uncapped demand for intelligence, AI embedded across economy, robotics, job disruption, potential cyber incident risk, and the economics of intelligence. It’s primarily narrative/strategic (few hard datapoints), but it supports a continued capex/compute buildout theme benefiting AI hardware, semis, networking, datacenters, and power/thermal infrastructure; with offsetting risks to labor-intensive s

Commentary advocating tighter U.S. policy to prevent Chinese companies from accessing AI/accelerator compute outside China (e.g., via third-country cloud/data centers), while allowing U.S. companies to use Chinese open-source software (OSS). No concrete policy action announced; it’s a directional regulatory/export-controls thesis.

Podcast-style discussion covering: (1) US policy/regulatory pressure around open-source AI vs closed models (Anthropic/OpenAI) and China model progress (Kimi K3); (2) a reported ~$1.5B Anthropic piracy/IP settlement (private company) and broader IP enforcement risk; (3) public-market reaction to surging AI capex with Google and Tesla cited as “tanking”; (4) NYC political rhetoric around evictions/property rights (potentially negative for exposed landlords/NYC CRE sentiment). Actionability is mod

Social-media discussion reacting to Sam Altman’s statement: he wants the US to “win in AI” across both open-source and proprietary models. No concrete policy, endorsement, funding, or company-specific catalyst is provided, so tradability is limited and mostly reinforces an existing pro-US AI leadership narrative.

Post claims Alphabet/Google noted in Q2 earnings remarks that demand for AI infrastructure is growing from robotics and “spatial intelligence” companies (including private company World Labs). This is a supportive data point for the AI infrastructure/compute/networking stack, but the source excerpt is light on numbers and not a direct, independently verifiable quote in this snippet.

Social posts claim AMD’s next-gen MI500 GPU platform may incorporate optical interconnects and be ahead of Nvidia’s Rubin Ultra in HBM, 4-die packaging, and scale-up domain. Separately, analyst Jeff Pu raises AI accelerator TAM to ~$1.4T by 2030 (from $1T) and lifts 2028 forecast to ~$1T; server CPU TAM >$220B by 2030 with “agentic AI” ~50% of TAM and discussion of CPU:GPU mix. This is high-level/rumor + sell-side TAM framing (directionally bullish for AI compute supply chain, but low verifiabil

Post cites informal channel checks that AMD’s Helios scale-up networking roadmap may adopt co-packaged optics (CPO) in future generations, and that broader hyperscaler/industry adoption of CPO is expected. This is a forward-looking packaging/interconnect thesis that could be bullish for AMD’s AI networking competitiveness and for public optics/photonics suppliers, but lacks specifics (names, timing, sourcing), reducing near-term tradability.